Anthropic proposes unsupervised algorithm ICM, matching or beating human-labeled training on several tasks
Original title:Unsupervised Elicitation
AI summary
Anthropic published on its Alignment Science Blog the unsupervised algorithm Internal Coherence Maximization (ICM), which elicits latent capabilities of pretrained models using only the model's own labels.
Quick Facts
- Published
- Updated
- Category
- Papers & Research
- AI score
- 66
原文介绍了 ICM 无监督算法的原理、基准结果与两条局限,读者可以据此评估无监督对齐路线目前的实际边界。
Related entities
Source & copyright
This article is based on curated titles and AI summaries from the AIHOT public API, with attribution and links back. Original copyright remains with the original publishers and authors; this site does not mirror full texts.
How to Cite
When citing this article, keep the title, date, and permalink, and credit Orange News as the source.
Orange News: Anthropic proposes unsupervised algorithm ICM, matching or beating human-labeled training on several tasks. June 1, 2025. https://new.nat6.net/en/news/ob15rh758b5krupo82xfq8qx2/